AI email personalization: five controls before you hit send
A personalized opening line can sound ready even when nobody’s sure where it came from. The recipient’s role may be outdated. A public post may belong to someone with the same name. The contact may have opted out since the draft was made.
Oscar Morgan’s framework for AI email personalization argues that a generated line should not reach the send queue until five things are visible: its source, its freshness, permission to use the underlying data, a traceable claim, and a named reviewer. Here is how each piece works.
️ Lock down the evidence boundary first
Before connecting a contact record to a drafting tool, the campaign owner should specify which fields are approved for this campaign and document which contacts are eligible to receive it. On a small team, one person may hold several of these roles. The decisions still need to be written down.
“Public” does not settle permission. The UK Information Commissioner’s Office notes that business contact details identifying an individual can be personal data, and using publicly available personal data for direct marketing still requires compliance with data protection rules. The ICO’s business-to-business marketing guidance also distinguishes corporate subscribers from sole traders and certain partnerships under UK electronic marketing rules.
Some material should not enter the drafting system at all. Scattered clues that suggest a person’s health, beliefs, financial trouble, or family situation should be excluded from outreach inputs entirely. The ICO’s guidance on special category data notes that profiling can create sensitive inferences from other information. Better to exclude those inputs than to ask a reviewer whether the resulting line sounds tactful.
Make every generated claim traceable
“Your team recently expanded into Europe” looks like a plausible opening line. If its source is a company page about an office that opened years ago, the word recently has no support. The sentence is fluent. The claim is not.
For each generated claim about the recipient or their organization, the framework calls for keeping the source link, retrieval date, relevant passage, generated wording, and review decision together. A reviewer can then inspect what the source actually establishes about that recipient at that time.
Unknown is a valid result. An unconfirmed role can be left out. An announcement with no clear date cannot be described as recent. Conflicting sources put the claim on hold until someone resolves them.
Traceability also matters after approval. If a source turns out to be wrong, the campaign owner needs to find every draft that relied on it. A link alone may lead to a page that has since changed, so the approved wording and the source state seen by the reviewer both need to be preserved.
Separate drafting from permission to send
A generated draft may be worth editing while the contact is still ineligible or a claim still needs proof. The framework recommends letting automation suggest copy, then requiring a named reviewer to release the exact version that will be sent. If either the recipient status or the wording changes, the earlier approval needs another look.
Risk determines who reviews. A verified company announcement fits an ordinary review. A new audience segment, an uncertain identity match, or a claim about an individual calls for the owner of that specific risk. A second approval cannot override an input policy that excludes a sensitive inference.
For US commercial email, the FTC’s CAN-SPAM guidance for businesses covers accurate sender and subject information, a valid postal address, an opt-out mechanism, and honoring opt-out requests. For teams sending to Gmail accounts, Google’s email sender guidelines describe unsubscribe expectations and additional requirements for bulk senders.
Test suppression before you scale
Before scaling, run the proposed workflow against a test set that includes: an opted-out contact whose profile otherwise looks ideal, a stale role, conflicting sources, a weak name match, a duplicate record under another address, and a case where the assigned reviewer is unavailable. Label the expected outcome for each record before running it.
- The opted-out contact must stay blocked even if a draft was previously approved.
- The stale role returns to data review.
- Conflicting sources hold the claim they affect.
- A weak name match cannot supply a personal opening.
- The duplicate must inherit the relevant suppression decision, or the same person re-enters the queue through another record.
- An unavailable reviewer leaves the message waiting with a visible owner.
Suppression needs a fresh check near the send decision, using the current record rather than the state captured when the draft was generated. Strong copy cannot compensate for a suppressed contact reaching the send queue.
Keep recovery human-owned
When a bad send happens, further sends using the same rule or source should pause while the campaign owner establishes how far the error traveled. The sent message, recipient set, source records, reviewer decision, and send time all need to be preserved before anyone repairs the inputs.
The data owner checks the faulty record and finds others sharing its source. The reviewer identifies the unsupported claim in queued drafts. The campaign owner determines which sends stay paused and whether a correction to recipients is appropriate. A sent email cannot be removed from every inbox, so recovery means stopping the faulty path and correcting the rule or data source before release resumes.
The framework closes with a straightforward test: if your team cannot show what supports a personal line, who cleared the contact, and who can stop the next send, that line should stay in draft.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
